摘要
Abstract
Deep learning-based side-channel attacks(DL-SCA)have emerged as a critical research direction in cryptographic security due to their capabilities in automated feature extraction and com-plex pattern learning.Compared to traditional SCA methods relying on manual feature engineering and statistical analysis,DL-SCA leverages deep neural networks to model nonlinear relationships be-tween physical leakage and secret operations,significantly improving attack efficiency and adaptability,particularly against noise,masking,and desynchronization.This paper presents a technical framework of DL-SCA,clarifying its general workflow and key phases.It systematically reviews mainstream model architectures,including multilayer perceptrons,convolutional neural network,recurrent neural network,generative adversarial network,and Transformer,along with their variants and recent ad-vancements.Applications of DL-SCA across power,electromagnetic,and cache-based side channels are analyzed.The study identifies critical challenges currently confronting DL-SCA,including model generalization,hyperparameter optimization,interpretability,robustness,and training cost efficiency.For each challenge,the study synthesizes and analyzes existing mitigation strategies from the literature.Finally,this study examines the future development trends of DL-SCA from two analytical dimen-sions:scenario-specific technological breakthroughs and the construction of a general-purpose technical framework.The study proposes concrete technical pathways and research recommendations for four paradigmatic application scenarios—small-sample contexts,multimodal systems,post-quantum cryp-tographic algorithms,and IoT devices—as well as cross-scenario general-purpose technologies.This systematic analysis aims to provide in-depth technical references centered on core technological inno-vations and their engineering implementations,thereby facilitating both theoretical advancements and practical applications of DL-SCA technology.关键词
侧信道攻击/深度学习/密码安全/神经网络模型/鲁棒性/强化学习/迁移学习Key words
side channel attack/deep learning/cryptographic security/neural network model/robustness/reinforcement learning/transfer learning分类
信息技术与安全科学
